modu-ai · GoosLab · Apr 3, 2026 · Apr 3, 2026 · Apr 3, 2026 · Apr 3, 2026
diff --git a/.moai/config/sections/system.yaml b/.moai/config/sections/system.yaml
@@ -27,9 +27,9 @@ github:
     enable_trust_5: true
     spec_git_workflow: feature_branch
 moai:
-    template_version: v2.9.0
+    template_version: v2.9.1
     update_check_frequency: daily
-    version: v2.9.0
+    version: v2.9.1
     version_check:
         cache_ttl_hours: 24
         enabled: true
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,68 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.9.1] - 2026-04-03
+
+### Summary
+
+Patch release fixing Agency v3.2 template deployment. All Agency files (agents, skills, commands, rules, config) are now correctly included in `moai update` distribution. Also includes CLAUDE.md v14.0.0 with Agency and Harness documentation, and README updates for all 4 languages.
+
+### Breaking Changes
+
+None
+
+### Fixed
+
+- Agency template deployment: 35 Agency files (agents, skills, commands, rules, .agency/ config) were missing from `internal/template/templates/` and not distributed via `moai update`
+
+### Changed
+
+- CLAUDE.md v14.0.0: Added /agency command reference, Agency agent catalog, Harness-Based Quality Routing, and Agency configuration section
+- README updated in 4 languages (en, ko, ja, zh) with Agency v3.2 details: GAN Loop, 5 specialized skills, 5-layer safety architecture, knowledge graduation, all 11 subcommands
+
+### Installation & Update
+
+```bash
+# Update to the latest version
+moai update
+
+# Verify version
+moai version
+```
+
+---
+
+## [2.9.1] - 2026-04-03 (한국어)
+
+### 요약
+
+Agency v3.2 템플릿 배포 누락 수정 패치. 모든 Agency 파일(에이전트, 스킬, 커맨드, 규칙, 설정)이 이제 `moai update`를 통해 올바르게 배포됩니다. CLAUDE.md v14.0.0 (Agency/Harness 문서) 및 4개 언어 README 업데이트 포함.
+
+### 주요 변경 사항 (Breaking Changes)
+
+없음
+
+### 수정됨 (Fixed)
+
+- Agency 템플릿 배포: 35개 Agency 파일(에이전트, 스킬, 커맨드, 규칙, .agency/ 설정)이 `internal/template/templates/`에 누락되어 `moai update`로 배포되지 않던 문제 수정
+
+### 변경됨 (Changed)
+
+- CLAUDE.md v14.0.0: /agency 커맨드 레퍼런스, Agency 에이전트 카탈로그, Harness 품질 라우팅, Agency 설정 섹션 추가
+- README 4개 언어(en, ko, ja, zh) Agency v3.2 상세 업데이트: GAN Loop, 5개 전문 스킬, 5계층 안전 아키텍처, 지식 졸업, 전체 11개 서브커맨드
+
+### 설치 및 업데이트 (Installation & Update)
+
+```bash
+# 최신 버전으로 업데이트
+moai update
+
+# 버전 확인
+moai version
+```
+
+---
+
 ## [2.9.0] - 2026-04-03
 
 ### Summary

diff --git a/CLAUDE.local.md b/CLAUDE.local.md
@@ -81,17 +81,45 @@ moai version            # Show version
 # All template changes MUST be made here
 internal/template/templates/.claude/
 internal/template/templates/.moai/
+internal/template/templates/.agency/
 internal/template/templates/CLAUDE.md
 ```
 
+### [HARD] Template-First Rule
+
+When adding new files to `.claude/`, `.moai/`, or `.agency/`:
+
+1. **Add to template FIRST**: `internal/template/templates/<path>`
+2. **Run `make build`** to regenerate embedded files
+3. **Then sync to local**: `moai update` or manual copy
+
+Never add files directly to the local project directories without also adding them to the template source. This includes:
+- New agents (`.claude/agents/`)
+- New skills (`.claude/skills/`)
+- New commands (`.claude/commands/`)
+- New rules (`.claude/rules/`)
+- New config files (`.moai/config/`)
+- New agency files (`.agency/`)
+
+**Verification**: Before committing, check that every new file under `.claude/`, `.moai/`, or `.agency/` has a corresponding file in `internal/template/templates/`.
+
 ### Local-Only Files (Never in Templates)
 ```
 .claude/settings.local.json    # Personal settings
+.claude/settings.json          # Rendered from .json.tmpl
+.claude/agent-memory/          # Per-project agent memory
+.claude/hooks/moai/handle-*.sh # Generated hook wrappers (not templates)
+.claude/commands/98-*.md       # Dev-project-specific commands
+.claude/commands/99-*.md       # Dev-project-specific commands
 CLAUDE.local.md                # This file
-.claude/hooks/moai/handle-*.sh  # Generated hook wrappers (not templates)
 .moai/cache/                   # Cache
 .moai/logs/                    # Logs
 .moai/state/                   # Session state storage
+.moai/specs/                   # Active SPEC documents
+.moai/plans/                   # Session plans
+.moai/reports/                 # Generated reports
+.moai/manifest.json            # Generated at runtime
+.moai/status_line.sh           # Rendered from .sh.tmpl
 ```
 
 ### Embedded Template System

diff --git a/CLAUDE.md b/CLAUDE.md
@@ -79,6 +79,17 @@ Default (natural language): Routes to autonomous workflow (plan -> run -> sync p
 
 Allowed Tools: Full access (Agent, AskUserQuestion, TaskCreate, TaskUpdate, TaskList, TaskGet, Bash, Read, Write, Edit, Glob, Grep)
 
+### Unified Skill: /agency
+
+Definition: Self-evolving creative production system for websites, landing pages, and web applications.
+
+Subcommands: brief, build, review, learn, evolve, resume, profile, phase, sync-upstream, rollback, config
+Default (natural language): Routes to agency pipeline (Planner -> Copywriter/Designer -> Builder -> Evaluator -> Learner)
+
+Pipeline: GAN Loop (Builder-Evaluator iterates up to 5 times until quality threshold 0.75 is met)
+
+For detailed Agency rules, see .claude/rules/agency/constitution.md
+
 ---
 
 ## 4. Agent Catalog
@@ -103,6 +114,14 @@ backend, frontend, security, devops, performance, debug, testing, refactoring
 
 agent, skill, plugin
 
+### Evaluator Agents (1)
+
+evaluator-active (independent skeptical quality assessment, 4-dimension scoring)
+
+### Agency Agents (6)
+
+planner, copywriter, designer, builder, evaluator, learner (self-evolving creative production pipeline)
+
 ### Dynamic Team Generation (Experimental)
 
 Agent Teams teammates are spawned dynamically using `Agent(subagent_type: "general-purpose")` with runtime parameter overrides from `workflow.yaml` role profiles. No static team agent definitions are used.
@@ -160,6 +179,18 @@ For team-based parallel execution of these phases, see .claude/skills/moai/team/
 
 For TRUST 5 framework details, see .claude/rules/moai/core/moai-constitution.md
 
+### Harness-Based Quality Routing
+
+MoAI-ADK uses a 3-level harness system for adaptive quality depth:
+
+- **minimal**: Fast validation for simple changes
+- **standard**: Default quality checks for most work
+- **thorough**: Full evaluator-active + TRUST 5 validation for complex SPECs
+
+Harness level is auto-determined by the Complexity Estimator based on SPEC scope. evaluator-active provides independent skeptical assessment with 4-dimension scoring (Functionality/Security/Craft/Consistency).
+
+**Configuration:** .moai/config/sections/harness.yaml, .moai/config/evaluator-profiles/
+
 ### LSP Quality Gates
 
 MoAI-ADK implements LSP-based quality gates:
@@ -267,6 +298,16 @@ MoAI-ADK uses Claude Code's official rules system at `.claude/rules/moai/`:
 - **Workflow rules**: Progressive disclosure, token budget, workflow modes
 - **Development rules**: Skill frontmatter schema, tool permissions
 - **Language rules**: Path-specific rules for 16 programming languages
+- **Agency rules**: AI Agency constitution (.claude/rules/agency/constitution.md)
+
+### Agency Configuration
+
+- `.agency/config.yaml`: Agency pipeline settings, adaptation weights, iteration limits
+- `.agency/context/`: Brand voice, visual identity, target audience, tech preferences
+- `.agency/fork-manifest.yaml`: Fork tracking for agency agents/skills evolved from MoAI upstream
+- `.moai/config/sections/constitution.yaml`: Project technical constraints (machine-readable)
+- `.moai/config/sections/harness.yaml`: Quality depth routing (minimal/standard/thorough)
+- `.moai/config/evaluator-profiles/`: Evaluator scoring profiles (default, strict, lenient, frontend)
 
 ### Language Rules
 
@@ -520,8 +561,8 @@ Large PDFs (>10 pages) return a lightweight reference when @-mentioned. Always s
 
 ---
 
-Version: 13.1.0 (Agent Teams Integration)
-Last Updated: 2026-03-29
+Version: 14.0.0 (Agency v3.2 + Harness Design Integration)
+Last Updated: 2026-04-03
 Language: English
 Core Rule: MoAI is an orchestrator; direct implementation is prohibited
 

diff --git a/internal/template/templates/.agency/config.yaml b/internal/template/templates/.agency/config.yaml
@@ -0,0 +1,83 @@
+# AI Agency v3.2 Configuration
+# Pipeline, GAN loop, evolution, and integration settings
+
+agency:
+  version: "3.2.0"
+  enabled: true
+
+  # Default tech stack
+  default_framework: "next.js"
+  design_tool: "variant"
+
+  # Pipeline configuration
+  pipeline:
+    phases: [planner, copywriter, designer, builder, evaluator, learner]
+    parallel_phases: [copywriter, designer]
+    team_mode_threshold: 6
+    default_mode: "step_by_step"
+
+  # GAN Loop (Builder <-> Evaluator)
+  gan_loop:
+    max_iterations: 5
+    pass_threshold: 0.75
+    escalation_after: 3
+    improvement_threshold: 0.05
+    strict_mode: false
+
+  # Self-evolution settings
+  evolution:
+    auto_evolve_threshold: 3
+    max_active_learnings: 50
+    archive_after_evolve: true
+    require_approval: true
+    max_evolution_rate_per_week: 3
+    cooldown_hours: 24
+    graduation_criteria:
+      minimum_observations: 5
+      minimum_confidence: 0.80
+      consistency_ratio: 0.80
+      staleness_window_days: 30
+
+  # Brand context
+  context:
+    dir: ".agency/context"
+    interview_on_first_run: true
+    update_interval: "monthly"
+
+  # Agent model assignments
+  models:
+    planner: opus
+    copywriter: sonnet
+    designer: sonnet
+    builder: sonnet
+    evaluator: sonnet
+    learner: opus
+
+  # Pipeline adaptation
+  adaptation:
+    enabled: true
+    confidence_threshold: 0.70
+    min_projects_for_adaptation: 5
+    phase_weights:
+      planner: 1.0
+      copywriter: 1.0
+      designer: 1.0
+      builder: 1.0
+      evaluator: 1.0
+      learner: 1.0
+    iteration_limits:
+      copywriter: 3
+      designer: 2
+      builder: 3
+
+  # User profile
+  profile:
+    user_style: "unknown"
+    projects_completed: 0
+    adaptation_confidence: 0.0
+
+  # moai integration
+  integration:
+    share_user_profile: true
+    worktree_isolation: true
+    upstream_sync_on_update: true
diff --git a/internal/template/templates/.agency/context/brand-voice.md b/internal/template/templates/.agency/context/brand-voice.md
@@ -0,0 +1,107 @@
+<!-- 
+  Agency Brand Context: Brand Voice
+
+  This file is part of the Agency Brand Context (constitution).
+  It is NOT auto-modified by the learner agent.
+  Changes require manual editing by the user.
+  It serves as the foundation for all agency skill evolution.
+
+  Fill this file during the first `/agency brief` run via client interview.
+-->
+
+# Brand Voice
+
+## Brand Name
+
+<!-- The official brand name as it appears in all communications. -->
+
+Name: _TBD_
+
+## Tagline
+
+<!-- A short, memorable phrase that captures the brand essence. -->
+
+Tagline: _TBD_
+
+<!-- Example: "Build better, ship faster." -->
+
+## Voice Attributes
+
+<!-- 3-5 adjectives that define how the brand sounds in every piece of copy. -->
+
+1. _TBD_
+2. _TBD_
+3. _TBD_
+
+<!-- Example: friendly, confident, approachable, knowledgeable, witty -->
+
+## Tone Spectrum
+
+<!-- Where does this brand sit on each spectrum? Mark with an X. -->
+
+Casual [ ] [ ] [ ] [X] [ ] Formal
+Playful [ ] [ ] [X] [ ] [ ] Serious
+Warm [ ] [ ] [ ] [X] [ ] Detached
+Bold [ ] [X] [ ] [ ] [ ] Reserved
+
+<!-- Adjust the X positions to reflect the brand's natural tone. -->
+
+## Language Rules
+
+Formality Level: _TBD_
+<!-- Example: semi-formal, conversational but not slangy -->
+
+Pronouns: _TBD_
+<!-- Example: "we" for the company, "you" for the customer, avoid "I" -->
+
+Jargon Policy: _TBD_
+<!-- Example: avoid technical jargon unless the audience is developers -->
+
+Sentence Length: _TBD_
+<!-- Example: short to medium, max 25 words per sentence -->
+
+Contractions: _TBD_
+<!-- Example: yes, always use contractions (don't, we're, it's) -->
+
+## Sample Phrases
+
+### Do
+
+<!-- Phrases that reflect the brand voice correctly. -->
+
+- _TBD_
+- _TBD_
+- _TBD_
+
+<!-- Example:
+- "Let's get you set up in under a minute."
+- "We've got your back, every step of the way."
+- "No fluff, just results."
+-->
+
+### Don't
+
+<!-- Phrases that violate the brand voice. -->
+
+- _TBD_
+- _TBD_
+- _TBD_
+
+<!-- Example:
+- "Dear valued customer, we wish to inform you..."
+- "HURRY! LIMITED TIME OFFER!!!"
+- "Per our previous correspondence..."
+-->
+
+## Emotional Tone
+
+<!-- What feeling should the copy evoke in the reader? -->
+
+Primary Emotion: _TBD_
+<!-- Example: confidence, trust, excitement, calm -->
+
+Secondary Emotion: _TBD_
+<!-- Example: curiosity, belonging, empowerment -->
+
+What the reader should think after reading: _TBD_
+<!-- Example: "These people understand my problem and have a real solution." -->